The Relevance Threshold field in the Documents settings refused negative values, so admins whose reranker returns raw scores (which can go below zero) could not set a useful threshold, and a negative RAG_RELEVANCE_THRESHOLD made the whole Documents settings page impossible to save while hybrid search was on. The server already stores and applies negative thresholds; only the field blocked them. The field now takes negative values in the same 0.01 steps as before, its help text says they are valid for rerankers that return raw scores outside the 0 to 1 range, and the hover hint asking for a value between 0 and 1 is gone. Nothing changes for 0 and above, and 0 still means no threshold.
Closes#29888
When a scheduled automation failed (for example because its model was deleted), the owner could only see it as an error in the run history, so they found out much later when the results stopped showing up. Automation run failed is now one of the automatic events users can pick for their notification targets under Settings > Notifications, and every failed run, scheduled or started by hand, sends the automation name and the error to those targets. In Admin Settings > General > Events, webhooks set to Specific users or groups now receive these failures when the owner is among those users or groups, and webhooks set to System events only stop receiving them.
Closes#33228
The workspace search showed its result labels such as "Personal chat" and "Personal note" in English for every interface language, and some of its own labels never reached the translation files. Several labels, prompts and banners in the Skills list, the skill editor and the terminal file browser were also hard-coded in English. They now follow the interface language. Folder and channel names in search results stay as the user typed them, except a folder named exactly like one of the five fixed labels, which shows translated as well, since the search results do not tell the two apart.
Fixes#33304Fixes#33305
Use "berkas" instead of "File" (108 existing uses vs 14), "Tambah" instead of
"Menambahkan" for the Add Files / Add Memory buttons (matching the other
"Add ..." actions), and "Pengaturan" instead of "Setelan". Also translate
"Enter certificate path", which was still in English.
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
When an MCP tool server could not be reached for a tool call (server restarted, connection reset by a proxy or load balancer, a request timing out), the whole chat was cancelled as if the user had pressed stop, and the log only said "Task was cancelled!". Now only that tool call fails, the model gets the connection error as the tool result and the reply finishes, and the underlying connection error is logged. The MCP SDK cancels whatever task opened the session when a single request fails, so each MCP session now runs in its own task instead of the chat's.
Fixes#33268
After changing the model or adding a second one to compare, the Web Search toggle showed the new default but the next message still went out with the previous model's setting. A model without Web Search as a default feature could search, and one with it did not. The message now follows what the toggle shows.
Fixes#33265
Remote fetches now ask for an uncompressed response: web pages, documents, remote images, profile pictures from a login provider and webhook deliveries. A response that comes back compressed anyway is refused, except for a webhook reply, which is only logged and is left unread instead.
A page or file whose host echoes a stored `Content-Encoding` regardless of what the client asks for (an S3 or MinIO object uploaded with that metadata) is no longer read, and a profile picture that arrives that way falls back to the default avatar.
When a foreground sub-agent stopped at the tool-call limit, the parent model got only the error, so everything the sub-agent had written before the limit was lost, while a background sub-agent in the same state reported it as partial output. A foreground sub-agent now returns the error followed by its partial output, the same way the background report does. The error in both modes is now the plain message, such as "Tool-call limit reached (30 iterations).", instead of raw internal data around it.
Fixes#33267
When the model handed a photo from the chat to a sub-agent, the call failed with "file_ids not attached or unavailable" and no sub-agent started. Text files and documents worked, only photos were refused. Photos attached in the chat are now accepted and reach the sub-agent as an image, the same way they reach the model in a normal chat. Files that were never attached to the chat are still refused.
Fixes#33241
Pressing Backspace inside an unconfirmed IME composition in the chat input ended the composition, so the next keystrokes started a new one and the text appeared twice. This started in 0.12.4, when the editor library picked up a release that treats every deletion as a Backspace key press, including deletions the IME makes inside a composition, and the editor's Backspace handling then disturbs the composition. Pinning that library to the last release without this behaviour restores normal IME editing.
Fixes#33223
A recent dev change marks a reply as having incomplete usage when the provider reports no token counts, so the session analytics can tell measured replies apart. The chat then treated that marker as usage and showed the Info button under every such reply, with a tooltip that only said "incomplete: true". The Info button now only appears when the reply carries actual usage data, while the stored marker stays as it is for analytics.
After flipping the Formatting switch in a note's menu, the editor stopped taking any input: typed text never appeared and nothing was saved, until the page was reloaded. The editor that comes back after the switch was still handed the one it replaced, and with the current editor library that old one now fails as soon as it is touched, so the new editor never connected to the note. Typing works right away after the switch now, and the text is saved and survives a reload.
Since a recent refactor on dev, the prompt and completion token counts of a reply that used tools added up every model call in the loop, even though each later call already re-sends the whole conversation. A reply whose context was 50k tokens showed 90k, the context gauge read over 100% and the next message triggered compaction long before the chat reached the threshold. Prompt and completion tokens show the latest call again, while total tokens and cost still add up across all calls.